On Tue, Apr 28, 2020 at 02:12:49PM +0000, Johannes Thumshirn wrote: > On 28/04/2020 13:28, Christoph Hellwig wrote: > > Instead of a convoluted chain just check for REQ_OP_READ directly, > > and keep all the memory stall code together in a single unlikely > > branch. > > Looks good, I just have one question, why the 'unlikely' annotation for > the psi code? > > In the current code, the psi path is not "unlikely" (only > REQ_OP_WRITE_SAME is). Is there any measurable benefit coming from the > annotation? Following the unlikely check in __bio_add_page that Johanness added there with the original addition.
